Uber's Journey in Building Multi-Modal Foundation Models

Explore Uber's innovative approach to developing in-house multi-modal foundation models in this 35-minute conference talk from Ray Summit 2024. Dive deep into Uber's strategy for automating and accelerating earner onboarding and screening processes across their global marketplace. Learn about the custom model architecture optimized for various input modalities and languages, and discover how these models were pretrained on Uber's vast proprietary data using distributed training clusters on Ray. Gain insights into the fine-tuning process and deployment of foundation models to streamline document processing workflows in key Uber markets globally. Understand how building scalable, high-performance foundation models using extensive enterprise data drives concrete business value in a global tech company setting. This talk by Ivan Zhou and Sabarish Sivanath from Uber offers valuable lessons on leveraging advanced AI technologies to enhance operational efficiency and user experience in large-scale digital platforms.